accounting

Nuvio 5.0 launches on Product Hunt, letting startups hand off invoicing, expenses, and reconciliation to an AI agent that can also connect with ChatGPT, Claude, or custom agents, plus new bookkeeping and tax services.

FinGREAT 评估开放权重 LLM 在金融图引导检索与关系分类任务上的表现，利用 11 节点 11 边的复式记账图谱分析 1300 份 SEC 10-K 文件，并引入包含 483 段人工标注的关系分类基准，最强开源与闭源模型准确率分别为 84.1% 和 92.8%。

A curated accounting skill library for AI finance agents covering close, reporting, reconciliation, AP, and AR workflows, built by analyzing prompts across finance-specific skill libraries.

Josh Pigford reports Intuit suspended his QuickBooks account after an AI agent helped categorize transactions, prompting a look at QuickBooks' terms of service, API limits, and five safer ways to connect AI agents to accounting data.

Elon Musk shared a post noting that AI models (which he jokingly calls 'Super Intelligence') now ace accounting tests, after falling short of the average accountant's ~37% score just eighteen months ago.

A tweet commenting on how frontier models like Claude have gone from being frowned upon to being effectively required on detail-oriented, well-specified work, citing a study where models outperformed even the best human accountants.

Mercor hired 12 licensed CPAs to complete simplified month-end close tasks from its APEX-Accounting benchmark and found that frontier AI models now score perfectly and are ~50x cheaper than the best junior accountant, having gone from below the human average to 100% accuracy in just 18 months.

Modus Audit Labs released Financial Audit Bench, an open-source benchmark testing whether AI can perform real junior-auditor work rather than merely answering accounting questions. Garry Tan highlighted the release, arguing every job will eventually have similar eval benchmarks.

Basis reports that using GPT-6 Astra has doubled the speed of completing a complex tax workbook and improved the efficiency and decision-making of their AI agents in accounting tasks.

The author argues that AI should automate routine tasks in middle management, HR, and accounting, enabling skilled workers to focus on productive work, while valuable managers who provide mentorship and expertise should remain.

Fina Business 在 Product Hunt 上线，推出面向创始人的 QuickBooks 替代方案，定价 20 美元/月，提供发票自动化、Stripe 支付、复式记账及财务报表等功能。

A company transformed an accounting firm to be AI native, increasing operating profit margins from 5% to 70% and is scaling this model to provide faster, cheaper accounting services to small businesses.

The article argues against concerns of an AI 'debt bomb' crisis, comparing off-balance-sheet financing for datacenters to historical practices like those used by Enron, suggesting scrutiny is warranted but panic is unwarranted.

Joanathan McIntosh, a 2x founder, launches Billow (YC S26), an AI-native accounting firm that replaces human accountants with agents to automate tasks in the $700B industry.

A practitioner reflects on building AI agents across multiple industries and finds that narrow, industry-specific, unglamorous use cases deliver real value, while broad 'AI runs the whole workflow' demos consistently underdeliver.

Palantir paid only £2.1m in UK corporation tax in 2024 despite £247m revenues, drawing criticism from unions and researchers over transfer pricing and tax avoidance practices.

APEX-Accounting, created by Ramp and Mercor, is a benchmark that tests AI models on 160 accounting scenarios to measure productivity.

The author shares their expense tracking setup using Hermes + Beancount + Feishu, leveraging the DeepSeek model for natural language bookkeeping. It offers reliable data validation, extremely low cost, and integration with Feishu documents.

APEX-Accounting is a benchmark created by Mercor and Ramp to assess frontier AI models on real accounting tasks. The best model, Claude-Fable-5 (Max), achieved 56.4% mean criteria.

Yohei Nakajima observes that society has repeatedly solved scalable collaboration with auditability using immutable logs, drawing parallels from open source (git), accounting, medicine, aviation, nuclear infrastructure, and scientific research. He suggests this pattern applies to long-running AI agent problems.